Flows and Cuts

Graph Theory (Fall 2011)

Rutgers University
Swastik Kopparty

1 Flows

We now study flows on graphs.

Definition 1 (Directed Graph). A directed graph G is a tuple (V, E) where E ⊆ V 2 . Here V is

the set of vertices and E is the set of directed edges. If (u, v) ∈ E, we say that there is an edge in
the graph G from u to v.
Further, if S, T ⊆ V , we denote by E(S, T ) the set of edges from some vertex in S to some vertex
in T :
E(S, T ) = {(u, v) ∈ E | u ∈ S, v ∈ T }.

Let G = (V, E) be a directed graph, and let s, t ∈ V . A flow in G from s to t is an allocation

of “current” to each edge of G, such that the net current entering any vertex v 6= s, t equals 0.
Formally, it is given by a function f : E → R such that f (e) ≥ 0 for each e ∈ E, and such that for
each v 6= s, t:
NetFlow(v) = 0,
where X X
NetFlow(v) = f (e) − f (e)
e∈E:e=(v,·) e∈E:e=(·,v)

(Notice that the expression for NetFlow(v) equals the total flow on edges out of v minus the total
flow on edges into v).
We have the following simple lemma.

Lemma 2. Let S ⊆ V . Let S denote V \ S. Then

X X X
NetFlow(v) = f (e) − f (e).
v∈S e∈E(S,S) e∈E(S,S)

Proof. Immediate from the definitions.

Applying this to S = V , we get that NetFlow(s)+NetFlow(t) = 0, and so NetFlow(s) = −NetFlow(t).

We call this quantity the value of the flow f from s to t.

Value(f ) = NetFlow(s).

1
2 Flows under capacity constraints

Let G = (V, E) be a directed graph. Let c : E → R be a function with c(e) ≥ 0 for each e ∈ E. We
will call c(e) the capacity of the edge E.
Let s, t ∈ V . A flow from s to t satisfying the capacities c is a flow f : E → R such that f (e) ≤ c(e)
for each edge e ∈ E.

3 Cuts

A cut in a graph G is simply a partition of the vertex set into two nonempty sets. If s, t are two
vertices of G, an (s, t)-cut is a partition of the vertex set into two nonempty sets such that s is in
one set and t is in the other.
Every cut in the graph G gives a simple upper bound on the maximum possible value of a flow
satisfying given capacity constraints.
Lemma 3. For every flow f satisfying the capacities c, and for every S ⊆ V , such that s ∈ S and
t ∈ S,
Value(f ) ≤ Capacity(S, S).

Proof. Using Lemma 2, we have:

X
NetFlow(s) = NetFlow(v)
v∈S
X X
= f (e) − f (e)
e∈E(S,S) e∈E(S,S)
X
≤ c(e) − 0
e∈E(S,S)

= Capacity(S, S).
Here we used the fact that 0 ≤ f (e) ≤ c(e) to derive the “≤” step.

4 Max-Flow / Min-Cut

In particular, the previous lemma implies that:

max Value(f ) ≤ min Capacity(S, S),
f S

where f varies over flows satisfying c, and S varies over (s, t)-cuts.
The max-flow-min-cut theorem says that these quantities are in fact equal.
Theorem 4 (Max-Flow/Min-Cut). Let G be a directed graph, and let c be a capacity function on
the edges of G. Then:
max Value(f ) = min Capacity(S, S),
f S

where f varies over flows satisfying c, and S varies over (s, t)-cuts.

2
The proof of this theorem will also lead to a simple, quick algorithm to find the maximum flow, as
well as a cut with capacity = value (thus showing that the flow is indeed max).

Proof. Let f be a flow satisfying c that maximizes Value(f ). We will show how to use f to find
a cut (S, S) whose capacity is Value(f ). Note that we crucially need to use the fact that f is a
maximum flow, this argument cannot work with any old flow.
The cut we are looking for is a “bottleneck” for f , across which no extra flow can be sent. Thus it
makes sense to look for the set S of vertices that can “receive” more flow. This is what we achieve
in the following procedure:

• Initialize S = {s}.

• Repeatedly do the following until S grows no further:

– If there exists u ∈ S and v ∈ V \ S are such that (u, v) ∈ E and either f(u,v) < c(u,v) or
f(v,u) > 0, then include v in S.

• Output the cut (S, S).

We first claim that (S, S) is an (s, t)-cut. If not, then t ∈ S. Consider the vertices v0 =
s, v1 , . . . , vk = t which led to t being included in S (i.e., for each i, either f(vi ,vi+1 ) < c(vi ,vi+1 )
or f(vi+1 ,vi ) > 0). Then we can modify f to get a flow with even higher value as follows: for a
suitably small , either increase f(vi ,vi+1 ) by  or decrease f(vi+1 ,vi ) by  (by choice of the vi , at least
one of these modifications will be possible without violating the capacity or nonnegativity of the
flow). This increases Value(f ) by , contradicting the maximality of f . Thus (S, S) is an (s, t)-cut.
Now we verify that Capacity(S, S) equals Value(f ).
We have:
X
Capacity(S, S) = c(e)
e∈E(S,S)
X
= f (e) by definition of S
e∈E(S,S)
X X
= f (e) − f (e) by definition of S again, f (e) = 0 for each e ∈ E(S, S)
e∈E(S,S) e∈E(S,S)

= Value(f ).

This completes the proof.

This leads to the following efficient algorithm to find a maximum flow.

• Start with f = 0.

• Keep doing the following until the max flow is found:

– Construct the set S following the algorithm in the proof of Theorem 4.

3
 – If (S, S) is an (s, t)-cut, then we are done; f is the desired max flow, and (S, S) is a cut
which proves that f is maximum.
– Otherwise, as in the proof of Theorem 4, we can modify f and increase Value(f ).

By inspecting the proof of Theorem 4 (and in particular paying attention to the amount by which
Value(f ) can be increased), it is easy to check that ifPall the capacities c(e) are integers, then this
algorithm will find the maximum flow with at most e c(e) modifications of f .
Another important corollary of this algorithm is the integrality of the max flow:

Corollary 5. If c(e) is an integer for each edge e ∈ E, then there is a maximum flow f where f (e)
is an integer for each edge e ∈ E.